# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/strands-decider-2b-a-small-open-source-decision-model-49987076

# Strands Decider 2B: a small, open\-source, decision model

240 points · 67 comments

[Full discussion](<https://news.ycombinator.com/item?id=49987076>)

[Read original](<https://strandsagents.com/blog/introducing-strands-decider/>)

Category: [Models & Products](<https://hacksnap.live/?category=models-products>)

## Skept-o-meter & Hotness

Skept\-o\-meter: Low\. Estimated from 6 comments\.

4 comments for the summary\.

Peak rank: \#3

Time in Top 10: 24\.0 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

63 recorded rank observations from 2026\-10\-07T09:02:38\.026986\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 63 recorded Hacksnap ranks:

2026\-10\-07T09:02:38\.026986\+00:00: rank \#8

2026\-10\-07T10:01:35\.069531\+00:00: rank \#8

2026\-10\-07T11:02:07\.770084\+00:00: rank \#8

2026\-10\-07T12:01:14\.996697\+00:00: rank \#8

2026\-10\-07T13:00:47\.45912\+00:00: rank \#7

2026\-10\-07T14:01:18\.307739\+00:00: rank \#7

2026\-10\-07T15:01:57\.541362\+00:00: rank \#5

2026\-10\-07T16:01:28\.474022\+00:00: rank \#4

2026\-10\-07T17:01:15\.036908\+00:00: rank \#4

2026\-10\-07T18:00:53\.300079\+00:00: rank \#3

2026\-10\-07T19:01:49\.456657\+00:00: rank \#3

2026\-10\-07T20:01:59\.032487\+00:00: rank \#4

2026\-10\-07T21:05:36\.655392\+00:00: rank \#6

2026\-10\-07T22:02:27\.105241\+00:00: rank \#6

2026\-10\-07T23:01:17\.452894\+00:00: rank \#5

2026\-10\-08T08:03:55\.083808\+00:00: rank \#4

2026\-10\-08T09:03:58\.68898\+00:00: rank \#116

2026\-10\-08T10:04:05\.364235\+00:00: rank \#117

2026\-10\-08T11:02:04\.410307\+00:00: rank \#116

2026\-10\-08T12:02:36\.39618\+00:00: rank \#116

2026\-10\-08T13:03:09\.206401\+00:00: rank \#116

2026\-10\-08T14:03:30\.113384\+00:00: rank \#117

2026\-10\-08T15:02:54\.444758\+00:00: rank \#115

2026\-10\-08T16:03:13\.067807\+00:00: rank \#114

2026\-10\-08T17:03:47\.92465\+00:00: rank \#113

2026\-10\-08T18:01:50\.360431\+00:00: rank \#113

2026\-10\-08T19:02:55\.719274\+00:00: rank \#114

2026\-10\-08T20:01:36\.359022\+00:00: rank \#114

2026\-10\-08T21:03:34\.000719\+00:00: rank \#116

2026\-10\-08T22:01:13\.459732\+00:00: rank \#116

2026\-10\-08T23:01:25\.318392\+00:00: rank \#115

2026\-10\-09T08:01:53\.314021\+00:00: rank \#116

2026\-10\-09T09:02:59\.93426\+00:00: rank \#117

2026\-10\-09T10:02:22\.652527\+00:00: rank \#116

2026\-10\-09T11:02:04\.807282\+00:00: rank \#116

2026\-10\-09T12:02:16\.769295\+00:00: rank \#116

2026\-10\-09T13:02:30\.35034\+00:00: rank \#118

2026\-10\-09T14:01:34\.215134\+00:00: rank \#118

2026\-10\-09T15:01:27\.564392\+00:00: rank \#118

2026\-10\-09T16:02:06\.561056\+00:00: rank \#119

2026\-10\-09T17:03:06\.673648\+00:00: rank \#121

2026\-10\-09T18:01:02\.851052\+00:00: rank \#121

2026\-10\-09T19:01:11\.250291\+00:00: rank \#122

2026\-10\-09T20:01:32\.91907\+00:00: rank \#122

2026\-10\-09T21:02:08\.376144\+00:00: rank \#124

2026\-10\-09T22:01:19\.253075\+00:00: rank \#126

2026\-10\-09T23:01:31\.990912\+00:00: rank \#126

2026\-10\-10T08:01:00\.593376\+00:00: rank \#125

2026\-10\-10T09:01:45\.527975\+00:00: rank \#127

2026\-10\-10T10:00:53\.799353\+00:00: rank \#127

2026\-10\-10T11:00:35\.65851\+00:00: rank \#126

2026\-10\-10T12:01:11\.169647\+00:00: rank \#126

2026\-10\-10T13:00:54\.008377\+00:00: rank \#125

2026\-10\-10T14:01:43\.208766\+00:00: rank \#127

2026\-10\-10T15:01:24\.825188\+00:00: rank \#127

2026\-10\-10T16:01:10\.213328\+00:00: rank \#126

2026\-10\-10T17:00:48\.84189\+00:00: rank \#125

2026\-10\-10T18:00:57\.55132\+00:00: rank \#125

2026\-10\-10T19:00:52\.379463\+00:00: rank \#124

2026\-10\-10T20:00:21\.993912\+00:00: rank \#124

2026\-10\-10T21:01:10\.657738\+00:00: rank \#125

2026\-10\-10T22:00:50\.433569\+00:00: rank \#124

2026\-10\-10T23:01:00\.95027\+00:00: rank \#125

Strands Decider 2B packages a 2B Qwen torso into a fast local decision model, but its JevBench gains and latency claims remain self\-reported; discussion asks about CPU, multimodal use and API naming\.

## The brief

Strands Agents released Strands Decider 2B, an open\-source decision model that strips the language\-model head from a Qwen3\.5\-2B torso and replaces it with a pointer head that scores predefined answer options\. It is designed for low\-latency, local binary or multiple\-choice decisions with confidence scores, not open\-ended generation or complex reasoning\. The team reports competitive accuracy and calibration on JevBench's public set and median latency around 115 ms on an RTX 3090, with weights, training data and scripts released on GitHub and Hugging Face\.

- Architecture: Qwen3\.5\-2B torso, LM head removed, pointer head of just over 1M parameters scores hidden states at option positions against the \<answer\> position; rank\-16 LoRA fine\-tune\.
- Performance: on JevBench's public set, 3rd of 33 in the 2B class and 1st of 30 excluding just\-over\-2B models; calibration measured with Brier score\.
- Latency: median around 115 ms on an RTX 3090 and around 153 ms on an M3 MacBook for small tasks; latency increases approximately linearly with task size\.
- Intended uses: model routing, tool selection, evaluations, guardrails, memory and context management, policy classification, and hybrid agents; includes a CLI and a Strands intervention example\.
- Limitations: decision models cannot generate text, are worse at complex problems, and are unsuited for coding, chatbots, document summarization and other common LLM tasks\.
- Release: open source on GitHub, weights on Hugging Face, with training data and scripts included\.

## Discussion themes

Analyzed: 2026\-10\-07T14:01:06\.281291\+00:00

Analysis sample: Based on 6 of 10 usable stored comments. Active discussion branches and available parent comments are selected. The analysis input was further shortened to fit its context limit.

This sample may omit parts of the full thread. Selected themes do not measure community opinion or how common a view is.

### Multimodal visual matching and calibrated probabilities

A commenter asks whether any of these models are multimodal yet and wants to try a model with calibrated probabilities on a shape\-matching question; a reply notes that Strands Decider can accept vision input and asks how it handles that question\.

Sources: [Comment 49988676](<https://news.ycombinator.com/item?id=49988676>) · [Comment 49988822](<https://news.ycombinator.com/item?id=49988822>)

### Origin and naming of \`noul\` for binary choices

One commenter questions why binary choices are called \`noul\`, asking whether it has meaning or just copies Jev's API; a reply attributes the term to Bernoulli\.

Sources: [Comment 49988832](<https://news.ycombinator.com/item?id=49988832>) · [Comment 49988852](<https://news.ycombinator.com/item?id=49988852>)

### Broad use\-case potential and request for examples

A commenter praises the article as unusually understandable for non\-experts, says the approach can technically be used for many use cases, and asks others to suggest ideas\.

Sources: [Comment 49987519](<https://news.ycombinator.com/item?id=49987519>)

### Human\-backed decision model as an alternative

A commenter jokes about a comedian releasing a decision model backed by humans, where a person named Jerry answers questions, as an alternative to the discussed model\.

Sources: [Comment 49987721](<https://news.ycombinator.com/item?id=49987721>)

## Sources & coverage

AI-generated summary · 2026\-10\-07T09:01:51\.639255\+00:00

Based on 4 of 4 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.